[clang] [ExprConstant] Handle shift overflow the same way as other kinds of overflow (PR #99579)

[Richard Smith via cfe-commits]

zygoloid wrote:

As noted in one of the inline comments, it doesn't seem good to be losing a diagnostic for ill-formed code. But this all makes sense to me as a step to make our behavior more consistent. I wonder if it'd make sense to make our folding behavior stricter by default as a prerequisite.

https://github.com/llvm/llvm-project/pull/99579